Automated classification of EEG signals in brain tumor diagnostics

In brain tumor diagnostics, EEG is most relevant in assessing how basic functionality is affected by the lesion and how the brain responds to treatments (e.g. post-operative). This paper focuses on developing an automated system to identify space-occupying lesions in the brain using EEG signals. We discuss major complications in relating EEG to different tumor classes and suggest an approach of feature extraction using wavelet techniques and classification by self-organizing maps. Initial tests show improvement over conventional frequency band features common in the EEG community. The tests also highlight the need to obtain efficient physically-motivated features as to how EEG is affected by various tumors